When the ADSM Client encounters a (valid) stale NFS handle in a filespace,
it is terminating backup of the entire parent filespace and reporting a
backup failure.  Why?  Is this intended behavior or a bug?  Is there an
option to change this behavior, or a reliable way to circumvent it?

We had two ADSM backup failures like this on the root filesystems ('/') of
different clients this week.  Both instances were caused by a transcient
mount of a remote filesystem (cdrom drive/contents) onto '/mnt'.  The cd
was probably later removed on the remote system, or that system was
rebooted or powered off.

This is not an uncommon occurrence on Unix systems.  Stale NFS handles
shouldn't be a big deal to ADSM, unless ADSM is expected (configured) to
back them up.  The default setting "All-Local" is supposed to exclude NFS
mounts anyway.  In our case, we use explicit 'domain' statements and are
not even interested in backing up NFS-mounted filesystems on our ADSM
clients.  We would like ADSM to just skip stale NFS handles and continue
backing up the rest of the parent filespace, and not report a failure.  We
could  minimize the problem by excluding commonly-used mountpoints like
'/cdrom' and '/mnt', but since Unix permits any arbitrary, unpredictable
directory (empty or not) to be used as a transcient mountpoint, we can't
eliminate all possible occurrences using the 'exclude' mechanism.

We're running ADSM Server v3r1, level 2.20 (Sun Solaris).  Clients are a
mix of Digital UNIX 4.0D/E, Sun Solaris 2.5.1/2.6/2.7, SGI Irix 6.5, HP-UX
10.20.  All clients are ADSM Client v3r1, level 0.6/0.7.

Kent Monthei
Data Recovery Services
SmithKline Beecham Pharmaceutical R&D